Attention arrives late

The pitch for spending money on a launch is that attention causes price. Buy the posts, buy the mentions, get the followers, and the chart follows. Everything sold to a new launch assumes that arrow points one way.

I had access to a dataset that can test it, and the arrow points the other way.

What the data is

Seventy-three crypto projects, tracked daily on CoinMarketCap for nine months from December 2025 to September 2026. For each project on each day: market capitalisation, follower count, and watchlist adds. Around eight thousand project-days in total, of which about half carry both a market cap and an attention figure.

These are ordinary listed projects, not memecoins, and not a random sample — they are the projects somebody was already paying attention to. Both of those limits matter and I come back to them.

The method

Seven-day changes rather than daily ones, because follower counts move slowly and a one-day difference is mostly noise. For every project and every date, the change in attention over the previous week, and the change in market cap over the previous week.

Then the same comparison at a series of offsets. Attention change this week against market cap change three days earlier. Against market cap change three days later. Against a week earlier, a week later, and so on out to three weeks in both directions.

If attention causes price, the correlation should be strongest looking forward: attention moves, then the chart moves. If price causes attention, it should be strongest looking backward.

The result

It is strongest looking backward, and it is not close.

Watchlist adds correlate with the market cap move from three days earlier at 0.224. With the market cap move three days later, the same figure is 0.028. Eight times weaker. By seven days out it has gone slightly negative.

Follower growth says the same thing more quietly: 0.148 looking back three days, −0.015 looking forward.

That asymmetry is the whole finding. People add a token to their watchlist after it has already moved. The attention is real, it is measurable, and it is downstream.

It is not an artefact of pooling

A correlation computed across many projects at once can be produced entirely by differences between them rather than by anything happening inside any one. So I ran it again separately for each project.

Of the fifty-seven projects with enough history to test, forty show price moving first. Ten show attention first. Seven show no clear direction. The strongest individual case reads 0.755 looking backward and −0.379 looking forward, which is about as clean a reversal as this kind of data produces.

What this does not show

The correlation is weak. An r of 0.22 accounts for about five per cent of the variation in market cap. Ninety-five per cent of what moves these charts is not in this measurement at all.

Correlation is not cause. Both series respond to the wider market, and a rising market moves prices and attention together without either driving the other.

CoinMarketCap surfaces coins that are already moving. Some of the watchlist adds that follow a price rise are people responding to a platform that showed them a riser, which is a mechanism of the site rather than a fact about human behaviour. I cannot separate those two with this data.

And the sample is projects a professional was already tracking. Whatever is true of them is not automatically true of a coin with nine holders and no listing.

The dataset is also not public. It is not mine to publish and it is not yours to check, which makes this the weakest-sourced thing on this site by some distance. Everything else here can be verified against a chain or an open API. This cannot. What I can publish is the method and the numbers it produced, and the method is simple enough to run against any similar dataset.

What it means for a launch

Two things, and the second is the useful one.

The first is that buying attention in order to create price is not the mechanism this data shows operating. It shows attention arriving after the move, on a three-day lag, consistently, across most of the projects tested. If that holds for a new launch — and it is a real if, given everything above — then a marketing spend timed to precede the move is paying for the wrong thing at the wrong time.

The second is that it agrees with what this desk found the hard way. Four launches, better names and cleaner ticker checks each time, and the one that earned anything earned it because an automated trader happened to be cycling that venue. That was a sample of four and it could easily have been luck. This is seventy-three projects over nine months arriving at the same shape from completely different data, and the two together are worth more than either on its own.

The conclusion I draw is narrow. Attention is worth having and it is worth measuring. It is not a lever you pull to start something, and a launch that is counting on it to be one is counting on the arrow pointing a direction this data says it does not point.